A precise force-position control method for servo electric actuators

Abstract

This invention provides a precision force-position control method for a servo electric actuator, comprising the following steps: S1: Constructing and training a two-input, three-output fuzzy neural network model for force control and a fuzzy neural network model for position control; S2: Constructing and training a two-input, two-output piecewise neural network model; S3: Obtaining the current force value and calculating the adjustment amount of the force controller parameters; S4: Obtaining the current position value and the adjustment amount of the position controller parameters; S5: Calculating the weighting coefficients of the force controller parameter adjustment amount and the position controller parameter adjustment amount, and adjusting the force controller parameter adjustment amount and the position controller parameter adjustment amount; S6: Adjusting the parameters of the force controller and the position controller, and calculating the force control amount and the position control amount respectively. This invention is used for force-position hybrid control of servo electric actuators to improve end-effector control accuracy.

Description

Technical Field

[0001] This invention relates to the field of servo electric actuator control technology, and in particular to a precision force and position control method for servo electric actuators. Background Technology

[0002] Servo electric actuators are an indispensable part of modern industrial automation control. Traditionally, servo electric actuator control involves position control, meaning movement along a predetermined trajectory. However, for applications requiring force control, position control alone is insufficient. Therefore, hybrid force-position control technology for servo electric actuators has emerged.

[0003] Currently, the application fields of servo electric actuator force-position hybrid control technology are very wide. For example, in the field of robotics, taking assembly robots as an example, CN106402080B discloses an integrated micro force-position hybrid servo hydraulic cylinder, which integrates displacement sensors, force sensors, and hydraulic cylinders, realizing modularity of force-position servo hydraulic control, improving control accuracy, reducing costs, and easily adapting to various hydraulically driven robot systems. However, in some high-requirement applications, the end-effector accuracy in the servo electric actuator force-position hybrid control process still suffers from insufficient accuracy.

[0004] In view of this, the present invention is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a precision force and position control method for servo electric actuators, which is used for force and position hybrid control of servo electric actuators to improve end-effector control accuracy.

[0006] To solve the above-mentioned technical problems, the present invention provides a precision force and position control method for a servo electric actuator, comprising the following steps:

[0007] S1: Construct and train a two-input, three-output fuzzy neural network model for force control and a fuzzy neural network model for position control. The inputs of the force control fuzzy neural network model are force deviation and force deviation rate of change, and the output is the parameter adjustment of the force controller. The inputs of the position control fuzzy neural network model are position deviation and position deviation rate of change, and the output is the parameter adjustment of the position controller.

[0008] S2: Construct and train a two-input two-output segmented neural network model. The inputs of the segmented neural network model are force deviation and position deviation, and the outputs are the weighted coefficients of the force-controlled fuzzy neural network model and the position-controlled fuzzy neural network model.

[0009] S3: Obtain the current force value and calculate the deviation and rate of change of the deviation between the current force value and the preset force target value, i.e., the force deviation and the rate of change of the force deviation; input the force deviation and the rate of change of the force deviation into the force control fuzzy neural network model trained by S1 to calculate the adjustment amount of the force controller parameters;

[0010] S4: Obtain the current position value and calculate the deviation and rate of change of deviation between the current position value and the preset target position value, i.e., the position deviation and the rate of change of position deviation; input the position deviation and the rate of change of position deviation into the position control fuzzy neural network model trained in S1 to calculate the adjustment amount of the position controller parameters;

[0011] S5: Input the force deviation calculated in S3 and the position deviation calculated in S4 into the segmented neural network model trained in S2, calculate the weighting coefficients of the force controller parameter adjustment and the position controller parameter adjustment, and adjust the force controller parameter adjustment and the position controller parameter adjustment.

[0012] S6: Adjust the parameters of the force controller and position controller respectively according to the adjustment amount of the force controller parameter and the adjustment amount of the position controller parameter after S5, and calculate the force control amount and the position control amount respectively.

[0013] Furthermore, the fuzzy neural network control model in S1 has four layers: an input layer, a fuzzification layer, a fuzzy inference layer, and an output layer.

[0014] The input layer consists of two nodes, corresponding to the deviation e and the rate of change of deviation ec, respectively.

[0015] The fuzzing layer has 2*n nodes. The node of the output layer deviation e is connected to the nodes 1 to n of the fuzzing layer, and the node of the output layer deviation change rate ec is connected to the nodes n+1 to 2n of the fuzzing layer, where n is the number of quantization levels of deviation e and deviation change rate ec.

[0016] The fuzzy inference layer has n*n nodes. The nodes 1 to n and n+1 to 2n of the fuzzification layer are paired and connected to each node of the fuzzy inference layer.

[0017] The output layer consists of three nodes, corresponding to △Kp, △Ki, and △Kd respectively. Each node in the output layer is connected to each node in the fuzzy inference layer.

[0018] Furthermore, the calculation formulas for each node in the input layer are as follows:

[0019] ;

[0020] Where x1 is e and x2 is ec;

[0021] The calculation formulas for each node of the blurring layer are as follows:

[0022] ;

[0023] Where, m 1b l 1b Let b = 1, 2, ..., n be the center and width of the membership function for each linguistic value of e, respectively, and m be the width of the membership function. 1b l 1b b = n+1, n+2, ..., 2n are the center and width of the membership function for each language value of ec, respectively;

[0024] The calculation formulas for each node in the inference layer are as follows:

[0025] ;

[0026] Wherein, b1 and b2 are two nodes in the fuzzification layer that are connected to node c in the inference layer;

[0027] The calculation formulas for each node in the output layer are as follows:

[0028] ;

[0029] Among them, w dc f4(1), f4(2), and f4(3) are the weights connecting each node in the inference layer to the node d in the output layer, respectively, where f4(1), f4(2), and f4(3) are △Kp, △Ki, and △Kd.

[0030] Furthermore, the loss value is calculated using the following formula when training the fuzzy neural network control model:

[0031] ;

[0032] Where g4(1), g4(2), and g4(3) are the expected outputs of △Kp, △Ki, and △Kd, respectively;

[0033] The fuzzy neural network control model is updated using the following formula during training: ab l ab w dc :

[0034] ;

[0035] ;

[0036] ;

[0037] Where n is the number of iterations, 0 < λ < 1, 0 < Υ < 1.

[0038] Furthermore, the segmented neural network model in S2 has three layers: an input layer, a hidden layer, and an output layer.

[0039] The input layer consists of two nodes, one for stress deviation e1 and the other for position deviation e2.

[0040] The hidden layer has m nodes, and each node in the input layer is connected to each node in the hidden layer, where 3 ≤ m ≤ 12;

[0041] The output layer consists of two nodes, which respectively weight the output coefficients α and β of the stress-controlled fuzzy neural network model and the position-controlled fuzzy neural network model. Each node in the hidden layer is connected to each node in the output layer; 0 < α < 1, 0 < β < 1.

[0042] Furthermore, when training the segmented neural network model, m is set to 3, 4, ..., 12 respectively for training; an evaluation function is set, and the value of the evaluation function is calculated for each of the m values ​​when training is 3, 4, ..., 12. The m corresponding to the maximum value of the evaluation function is set as the number of nodes in the hidden layer of the segmented neural network model. The form of the evaluation function is as follows:

[0043] ;

[0044]

[0045] Where j is the number of evaluation indicators, θ i The sum of F is 1. i F is the value of the preset evaluation index i. i,min F is the minimum value of the preset evaluation index i during training for each m value. i,max This is the maximum value of the preset evaluation index i during training for each m value.

[0046] Furthermore, before training the segmented neural network model, θ is obtained through the following steps. i :

[0047] A1: Construct a two-dimensional expert evaluation table, with each behavior representing θ. i Each column is assigned a importance level, and values ​​are assigned to each importance level from highest to lowest.

[0048] A2: Obtain the expert evaluation form and calculate the total score H(θ) for each indicator. i ), and calculate θ using the following formula i :

[0049] .

[0050] This invention provides a precision force and position control method for a servo electric actuator, which has the following advantages: A fuzzy neural network model for force control and a fuzzy neural network model for position control are constructed to calculate the parameter adjustment amounts of the force controller and the position controller. Furthermore, a piecewise neural network model is constructed to comprehensively calculate the weighting coefficients α and β of the force controller parameter adjustment amount and the position controller parameter adjustment amount based on force deviation and position deviation. Finally, based on the parameter adjustment amounts of the force controller and the position controller, as well as the weighting coefficients α and β, the parameters of the force controller and the position controller are determined to define the force control amount and the position control amount, thereby improving the end-effector control accuracy. Attached Figure Description

[0051] Figure 1 This is a control flowchart of a precision force and position control method for a servo electric actuator, as shown in the embodiment. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Example 1:

[0054] A method for precision force and position control of a servo electric actuator includes the following steps:

[0055] S1: Construct and train a two-input, three-output fuzzy neural network model for force control and a fuzzy neural network model for position control. The inputs to the force control fuzzy neural network model are force deviation and the rate of change of force deviation, and the output is the parameter adjustment of the force controller. The inputs to the position control fuzzy neural network model are position deviation and the rate of change of position deviation, and the output is the parameter adjustment of the position controller.

[0056] S2: Construct and train a two-input, two-output segmented neural network model. The inputs of the segmented neural network model are force deviation and position deviation, and the outputs are the weighted coefficients of the force-controlled fuzzy neural network model and the position-controlled fuzzy neural network model.

[0057] S3: Obtain the current force value and calculate the deviation and rate of change of the deviation between the current force value and the preset force target value, i.e., the force deviation and the rate of change of the force deviation; input the force deviation and the rate of change of the force deviation into the force control fuzzy neural network model trained by S1 to calculate the adjustment amount of the force controller parameters;

[0058] In this embodiment, the force controller can be configured as a PID controller, and the initial parameters Kp, Ki, and Kd of the force controller can be preset. During the control process, the force deviation and the rate of change of the force deviation are input into the force control fuzzy neural network model trained by S1 to calculate the adjustment amount of the force controller parameters ΔKp, ΔKi, and ΔKd.

[0059] S4: Obtain the current position value and calculate the deviation and rate of change of deviation between the current position value and the preset target position value, i.e., the position deviation and the rate of change of position deviation; input the position deviation and the rate of change of position deviation into the position control fuzzy neural network model trained in S1 to calculate the adjustment amount of the position controller parameters;

[0060] In this embodiment, the position controller can be configured as a PID controller, and the initial parameters Kp, Ki, and Kd of the position controller can be preset. During the control process, the position deviation and the rate of change of position deviation are input into the position control fuzzy neural network model trained by S1 to calculate the adjustment amount of the position controller parameters ΔKp, ΔKi, and ΔKd.

[0061] S5: Input the force deviation calculated in S3 and the position deviation calculated in S4 into the segmented neural network model trained in S2, calculate the weighting coefficients of the force controller parameter adjustment and the position controller parameter adjustment, and adjust the force controller parameter adjustment and the position controller parameter adjustment.

[0062] S6: Adjust the parameters of the force controller and position controller respectively according to the adjustment amount of the force controller parameter and the adjustment amount of the position controller parameter after S5, and calculate the force control amount and the position control amount respectively.

[0063] In this embodiment, in step S5, the weighting coefficient α and the weighting coefficient β of the force controller parameter adjustment amount are calculated based on the force deviation and position deviation. In step S5, the force controller parameter adjustment amounts △Kp, △Ki, and △Kd obtained in step S3 are multiplied by α to adjust the force controller parameters. Then, in step S6, the preset initial parameters Kp, Ki, and Kd of the force controller are added to obtain the final parameters of the force controller, and the force control amount is calculated. In step S5, the position controller parameter adjustment amounts △Kp, △Ki, and △Kd obtained in step S4 are multiplied by β to adjust the position controller parameters. Then, in step S6, the preset initial parameters Kp, Ki, and Kd of the position controller are added to obtain the final parameters of the position controller, and the force control amount is calculated.

[0064] This embodiment of a precision force-position control method for a servo electric actuator constructs a force control fuzzy neural network model and a position control fuzzy neural network model to calculate the parameter adjustment amounts of the force controller and the position controller. Furthermore, a segmented neural network model is constructed to comprehensively calculate the weighting coefficients α and β of the force controller parameter adjustment amounts based on force deviation and position deviation. Finally, based on the parameter adjustment amounts of the force controller and the position controller, as well as the weighting coefficients α and β, the final parameters of the force controller and the position controller are determined to define the force control amount and position control amount, thereby improving the end-effector control accuracy.

[0065] In an optional embodiment, the fuzzy neural network control model in S1 has four layers: an input layer, a fuzzification layer, a fuzzy inference layer, and an output layer.

[0066] The input layer consists of two nodes, corresponding to the deviation e and the rate of change of deviation ec, respectively.

[0067] The fuzzing layer has 2*n nodes. The node of the output layer deviation e is connected to the nodes 1 to n of the fuzzing layer, and the node of the output layer deviation change rate ec is connected to the nodes n+1 to 2n of the fuzzing layer, where n is the number of quantization levels of deviation e and deviation change rate ec.

[0068] The fuzzy inference layer has n*n nodes. The nodes 1 to n and n+1 to 2n of the fuzzification layer are paired and connected to each node of the fuzzy inference layer.

[0069] The output layer consists of three nodes, corresponding to △Kp, △Ki, and △Kd respectively. Each node in the output layer is connected to each node in the fuzzy inference layer.

[0070] In this optional embodiment, combining fuzzy algorithms and RBF neural networks to construct a fuzzy neural network model can improve the control response speed, stability, and accuracy.

[0071] Where n is the number of linguistic values ​​when the deviation e and the deviation change rate ec are fuzzified. For example, if n=7, the linguistic values ​​are NB, NM, NS, Z, PS, PM, and PB. At this time, there are 14 nodes in the fuzzification layer and 49 nodes in the fuzzy inference layer. Nodes 1 to 7 of the fuzzification layer are connected to the nodes of the input layer e, and nodes 8 to 14 are connected to the nodes of the input layer ec. Nodes 1 to 7 and nodes 8 to 14 of the fuzzification layer are paired and connected to the 49 nodes of the fuzzy inference layer.

[0072] In an optional embodiment, the calculation formula for each node of the input layer is as follows:

[0073] ;

[0074] Where x1 is e and x2 is ec;

[0075] The calculation formulas for each node of the blurring layer are as follows:

[0076] ;

[0077] Where, m 1b l 1b Let b = 1, 2, ..., n be the center and width of the membership function for each linguistic value of e, respectively. In practice, the center and width of the membership function for each linguistic value can be set as needed, which will not be elaborated here. m 1b l 1b b = n+1, n+2, ..., 2n are the center and width of the membership function for each language value of ec, respectively;

[0078] The calculation formulas for each node in the inference layer are as follows:

[0079] ;

[0080] Wherein, b1 and b2 are two nodes in the fuzzification layer that are connected to node c in the inference layer;

[0081] The calculation formulas for each node in the output layer are as follows:

[0082] ;

[0083] Among them, w dc f4(1), f4(2), and f4(3) are the weights connecting each node in the inference layer to the node d in the output layer, respectively, where f4(1), f4(2), and f4(3) are △Kp, △Ki, and △Kd.

[0084] In an optional embodiment, the loss value is calculated using the following formula when training the fuzzy neural network control model:

[0085] ;

[0086] Where g4(1), g4(2), and g4(3) are the expected outputs of △Kp, △Ki, and △Kd, respectively;

[0087] The fuzzy neural network control model is updated using the following formula during training: ab l ab w dc :

[0088] ;

[0089] ;

[0090] ;

[0091] Where n is the number of iterations, 0 < λ < 1, 0 < Υ < 1.

[0092] In an optional embodiment, the segmented neural network model in S2 has three layers: an input layer, a hidden layer, and an output layer.

[0093] The input layer consists of two nodes, one for stress deviation e1 and the other for position deviation e2.

[0094] The hidden layer has m nodes, and each node in the input layer is connected to each node in the hidden layer, where 3 ≤ m ≤ 12;

[0095] The output layer consists of two nodes, which respectively weight the output coefficients α and β of the stress-controlled fuzzy neural network model and the position-controlled fuzzy neural network model. Each node in the hidden layer is connected to each node in the output layer; 0 < α < 1, 0 < β < 1.

[0096] In this optional embodiment, a segmented neural network model is constructed using a BP neural network to comprehensively calculate the weighting coefficient α of the force controller parameter adjustment and the weighting coefficient β of the position controller parameter adjustment based on the force deviation and position deviation.

[0097] In an optional embodiment, when training the segmented neural network model, m is set to 3, 4, ..., 12 respectively for training; an evaluation function is set, and the value of the evaluation function is calculated for each of the m values ​​being 3, 4, ..., 12. The m corresponding to the maximum value of the evaluation function is set as the number of nodes in the hidden layer of the segmented neural network model. The evaluation function has the following form:

[0098] ;

[0099]

[0100] Where j is the number of evaluation indicators, θ i The sum of F is 1. i F is the value of the preset evaluation index i. i,min F is the minimum value of the preset evaluation index i during training for each m value. i,max This is the maximum value of the preset evaluation index i during training for each m value.

[0101] In this optional embodiment, the range of m values ​​(i.e. the number of hidden layer nodes) is first determined, and different m values ​​are used to train the segmented neural network model. Then, the optimal m value is determined based on the evaluation function after training, which is used as the final number of hidden layer nodes in the segmented neural network model.

[0102] In an optional embodiment, θ is obtained before training the segmented neural network model through the following steps. i :

[0103] A1: Construct a two-dimensional expert evaluation table, with each behavior representing θ. i Each column is assigned a importance level, and values ​​are assigned to each importance level from highest to lowest.

[0104] A2: Obtain the expert evaluation form and calculate the total score H(θ) for each indicator. i ), and calculate θ using the following formula i :

[0105] .

[0106] In this optional embodiment, the importance level of A1 can be very important, moderately important, or not important, corresponding to 3 points, 2 points, and 1 point, respectively. After constructing the two-dimensional expert evaluation table, 10 experts can be invited to evaluate the importance of each indicator. Based on the experts' evaluations and the scores corresponding to the importance level in A2, the total score of each indicator is calculated, and finally, the θ corresponding to each indicator is calculated. i In summary, this optional embodiment can quickly, effectively, and accurately obtain θ corresponding to each indicator. i .

[0107] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for precise force and position control of a servo electric actuator, characterized in that, Includes the following steps: S1: Construct and train a two-input, three-output fuzzy neural network model for force control and a fuzzy neural network model for position control. The inputs of the force control fuzzy neural network model are force deviation and force deviation rate of change, and the output is the parameter adjustment of the force controller. The inputs of the position control fuzzy neural network model are position deviation and position deviation rate of change, and the output is the parameter adjustment of the position controller. S2: Construct and train a two-input two-output segmented neural network model. The inputs of the segmented neural network model are force deviation and position deviation, and the outputs are the weighted coefficients of the force-controlled fuzzy neural network model and the position-controlled fuzzy neural network model. S3: Obtain the current force value and calculate the deviation and rate of change of the deviation between the current force value and the preset force target value, i.e., the force deviation and the rate of change of the force deviation; input the force deviation and the rate of change of the force deviation into the force control fuzzy neural network model trained by S1 to calculate the adjustment amount of the force controller parameters; S4: Obtain the current position value and calculate the deviation and rate of change of deviation between the current position value and the preset target position value, i.e., the position deviation and the rate of change of position deviation; input the position deviation and the rate of change of position deviation into the position control fuzzy neural network model trained in S1 to calculate the adjustment amount of the position controller parameters; S5: Input the force deviation calculated in S3 and the position deviation calculated in S4 into the segmented neural network model trained in S2, calculate the weighting coefficients of the force controller parameter adjustment and the position controller parameter adjustment, and adjust the force controller parameter adjustment and the position controller parameter adjustment. S6: Adjust the parameters of the force controller and position controller respectively according to the adjustment amount of the force controller parameter and the adjustment amount of the position controller parameter after S5, and calculate the force control amount and the position control amount respectively; The segmented neural network model in S2 has three layers: an input layer, a hidden layer, and an output layer. The input layer consists of two nodes, one for stress deviation e1 and the other for position deviation e2. The hidden layer has m nodes, and each node in the input layer is connected to each node in the hidden layer, where 3 ≤ m ≤ 12; The output layer consists of two nodes, which respectively weight the output coefficients α and β of the stress-controlled fuzzy neural network model and the position-controlled fuzzy neural network model. Each node in the hidden layer is connected to each node in the output layer; 0 < α < 1, 0 < β < 1. When training the segmented neural network model, m is set to 3, 4, ..., 12 respectively for training; an evaluation function is set, and the value of the evaluation function is calculated for each of the m values ​​of 3, 4, ..., 12. The m corresponding to the maximum value of the evaluation function is set as the number of nodes in the hidden layer of the segmented neural network model. The form of the evaluation function is as follows: ； Where j is the number of evaluation indicators, θ i The sum of F is 1. i F is the value of the preset evaluation index i. i,min F is the minimum value of the preset evaluation index i during training for each m value. i,max This is the maximum value of the preset evaluation index i during training for each m value.

2. The precision force and position control method for a servo electric actuator according to claim 1, characterized in that, The fuzzy neural network control model in S1 has four layers: an input layer, a fuzzification layer, a fuzzy inference layer, and an output layer. The input layer consists of two nodes, corresponding to the deviation e and the rate of change of deviation ec, respectively. The fuzzing layer has 2*n nodes. The node of the output layer deviation e is connected to the nodes 1 to n of the fuzzing layer, and the node of the output layer deviation change rate ec is connected to the nodes n+1 to 2n of the fuzzing layer, where n is the number of quantization levels of deviation e and deviation change rate ec. The fuzzy inference layer has n*n nodes. The nodes 1 to n and n+1 to 2n of the fuzzification layer are paired and connected to each node of the fuzzy inference layer. The output layer consists of three nodes, corresponding to △Kp, △Ki, and △Kd respectively. Each node in the output layer is connected to each node in the fuzzy inference layer.

3. The precision force and position control method for a servo electric actuator according to claim 2, characterized in that, The calculation formulas for each node in the input layer are as follows: ； Where x1 is e and x2 is ec; The calculation formulas for each node of the blurring layer are as follows: ； Where, m 1b l 1b Let b = 1, 2, ..., n be the center and width of the membership function for each linguistic value of e, respectively, and m be the width of the membership function. 1b l 1b b = n+1, n+2, ..., 2n are the center and width of the membership function for each language value of ec, respectively; The calculation formulas for each node in the inference layer are as follows: ； Wherein, b1 and b2 are two nodes in the fuzzification layer that are connected to node c in the inference layer; The calculation formulas for each node in the output layer are as follows: ； Among them, w dc f4(1), f4(2), and f4(3) are the weights connecting each node in the inference layer to the node d in the output layer, respectively, where f4(1), f4(2), and f4(3) are △Kp, △Ki, and △Kd.

4. The precision force and position control method for a servo electric actuator according to claim 3, characterized in that, The loss value is calculated using the following formula when training the fuzzy neural network control model: ； Where g4(1), g4(2), and g4(3) are the expected outputs of △Kp, △Ki, and △Kd, respectively; The fuzzy neural network control model is updated using the following formula during training: ab l ab w dc : ； ； ； Where n is the number of iterations, 0 < λ < 1, 0 < Υ < 1.

5. The precision force and position control method for a servo electric actuator according to claim 1, characterized in that, Before training the segmented neural network model, θ is obtained through the following steps. i : A1: Construct a two-dimensional expert evaluation table, with each behavior representing θ. i Each column is assigned a importance level, and values ​​are assigned to each importance level from highest to lowest. A2: Obtain the expert evaluation form and calculate the total score H(θ) for each indicator. i ), and calculate θ using the following formula i : 。

Images